inconfident2021

> It turned out that architectural improvements really did matter

Transformer (2017). NET based Diffusion (2015). Score based diffusion (2019). DDPM (2020). Uses unet (2015). Clip (~2021) uses Resnet (2015) + ViT (2020). stable diffusion also uses Unet.

Yes, deep learning is easy. Throw compute, you get the answer.

Those with insights on the old problem are now churning papers because they throw compute and deep learning to the stuffs they understood well.

You can look into any paper and see inspiration from old methods. Applying deep learning is hard, deep learning itself is quite easy. If it were really hard, it wouldn't have been popular at all! Nowadays most of the people don't even bother with architectural gains.

Sure if better architecture eventually comes up, people will throw new ones into the compute. I don't believe in bayesian stuff either. However it is worth learning otherwise you might miss insights on a lot of papers!

Whatever you do, just having the idea to understand problem matters a lot. Then later comes the part of deep learning and throwing compute.

People say our job is safe, but aren't we in similar position to the horses who got replaced by cars or telegraph operators who just vanished?

There seems to be many people who thinks sales is the way to go. So does it mean that being a average engineer will have no value in substantial future?

How can one demonstrate this skill? When academics look for grades and HR looks for buzz word. Is there some keywords that people are missing which will help them demonstrate it?

I agree with public speaking. Whenever I see someone able to even create a video and upload with a face, I am in awe with their confidence.

Public speaking just takes everything to the next level. How can people be so confident about themselves?

For example some technical blogger post it like they own their posts. Going through that feels like the author is a all-knowing being.

How does that even kick in within a person?

I call this survivorship bias. No matter what you make, only few care enough. Find this few is hardest part. Because there are millions others doing the same damn thing.

Everyone has a kick for the survivorship bias. At the end it is just that.

How about other who made similar apps? How about all those aspiring game devs who actually put games, web developers who launched products or even some guys who made something apple would want?

The OP managed to get lucky.

You can now read new papers and follow ups from the community! It is really useful. You already know chain of thoughts. So you are on a good track!

Try to use these LLMs to automate more of the mundane tasks. Like scripting in bash, compiling videos, converting documents, refactoring data, transpiling code, transcribing code, etc. You will begin to see what works and what doesn't!

At the same time try to come with fun challenges for yourself to fool it. That will aid in learning ways to make it more obedient to you.

They would probably not pass in the div or filler elements.

The reason I passed the div was I wanted things to be surrounded in that exact space. So when the model output it, the button would be in right place with right size.

The extra filler forms a guiding factor that helps will be stored in the context! I did that using GPT3.5.

This is a good list of points! Knowingly or unknowingly I have been doing most of these things. 4-5 is currently unreachable for me. Others are quite manageable!

I have zero design skills and I had to create something using javascript. I gave some prompts and it was able to come up with a pop-up box. It needed some tweaking but having to not write all of that was really satisfying.

https://chat.openai.com/share/cb3a477b-57bd-46fd-92c9-4a3016...

I have attached the example in the above chat.

You write requirements and your expectations and make the model match your expectations. Until you have clear expectations of what you want from it, prompting LLM is pretty useless. It cannot do highly specific task because those are limited in the original training corpus too. However, for more generic task, it has seen most of the stuff out there, so it should be good enough. Having clarity on your problem is the key.

You have to make sure to couple chain of thought with branching, analysis and evaluation, then you can get pretty good results.

> understand X number of words, or pages, or paragraphs.

LLMs do not have the ability to reason with numbers. Most of the time they are hallucinating. One good strategy is to make it output in list and define the structure for each item of the list. If you give an example of what your list should look like, it will give you something close it.

> has trouble carrying a list forward without losing items.

This is the fundamental problem with these models because of the context limit. When you are prompting always remember that is processing a huge paragraph and emitting the next sentences of the paragraph. If you want information to be carried onwards, you have make it output on every prompt or you can also try to use specific identifiers. LLMs are good at in-context learning. It will not work 100% of the time, but it is usually good than having nothing at all.

> I want to speak chinese to the AI and I want it to correct me.

Give it a role of tutor and describe the instructions what the tutor should do.

I don't know what I can say about others, as for me, I am quite good at getting what I want from LLMs. I created a chat app using chatgpt without ever having background in frontend web development. I managed to do it in 2 days. Think of going from 0 to fully functional client in 2 days without having any idea about frontend.

It's a prototype. It doesn't have many states. But it gets the job done!

I have different other experiences where prompting has helped me. The only issue is I think it will be irrelevant in future given how fast these models are improving.

> at the very least, if you've gone so long and have gotten N hundred rejections you know it probably won't work and you'll have to make alternate plans.

I will start preparing for this! As an alternative I am thinking of PhD but that requires good grades in Master and I messed it up. So I have to think backup of backup!

My only question, why aren't they using gltf? What is the issue with gltf? It supports most of the things?

I understand! Thank you so much for the execution plan. Yes, I am CS major who is about to graduate soon.

I will review my resume and try to make it better! Figuring out the rejection is the hardest part.

The framework looks solid! My pressing issue is I will not have stable source of income in near future.

The sad episode kicked in because of the current market. I have no experience to show, no remarkable achievements, no cool projects, and everything is vanilla. This process spiraled downward to current stage.

Also graduation is taking longer than I expected.

I haven't started my career yet. :(

Ruined bachelors. Ruined masters. Mid profile. No potential to be hired. Zero confidence in skills. Bad resume. Bad language skills. Horrible ego.

I myself am a happy person. I can enjoy watching the void. Seriously, I can waste time like that and be completely happy.

But it makes me kinda sad when I realize that I need to be responsible. Because of the external factor it forces me out of my comfort. As a result, these external stuffs makes me regret the time I enjoyed when I was staring to the void.

Even for AI/ML mine is restricted to vision/graphics. I have some experience with classical ML methods but they dwarf the experience in deep learning based methods. Even there most of my skills are around prototyping and research. I don’t have confidence to apply at all. Therefore I wanna transition to something manageable. The skills at the most venues are really beyond what I have.

I am not being humble, I am just saying I suck quite bad at it.

I started blogging a while ago, but all my contents are mid at best. How do you level up? There is constant pressure within me to produce but I can't write anything impressive that can help other people. All the things I do rely on collective knowledge of the common, hence I will not be adding anything new on the table. This sabotages my self confidence. Are you in similar situation too? I am in that 90% of the internet where everything is crap.

I can’t comment or argue with prominent scientists such as LeCunn. But I surely can give perspective as SWE about to enter the industry.

1. We live in an uncertain time. Before most of the people, at least with a bit of research could have predicted the trajectory of technological innovation. Being born in era of tech, it was very easy to adopt to it. From PC to now, I have embraced every technological progress. Yet, nothing sent chills like this one.

2. I do not worry about my job. What I worry is most of the world is about getting shit done and this tech helps you get your shit done (for now). If we are at the upper tail of some sigmoid, there is nothing to fear. If we are at slope, oh boy, I cannot fathom the economic domino this tech is about to bring out. People are not rational. We are mostly governed by emotion. If you think, every single joe is rational and has capacity to judge, nobody would have voted from Trump to be president or riot the Capitol.

3. Currently it can do 60% of the work. If it evolves, it will do 80% of the work. If you factor human population and the amount of inefficiency in current economy, this will create a boom of productive workers with bust of average worker. And the world is not just for elites. Average people need their livelihood too. Now think about the magnitude of displacement.

4. I have no issue with innovation and breakthrough. AI won’t kill us, because we will kill each other before that. This tech is exposing entire middle class job (check out the job being exposed. There is a paper for that from openAI titled: gpts are gpts) And there are lot if people in that bracket. For once, think about us too.

So, these religious people do not exclusively worry about immediate effects of the tech. But within their doom, hides an issue that needs to be heard.

That’s my two cents.

I hate to say this but symbolic reasoning and planning has aged like milk. The use case is so limited and the only people using these are the people full of nostalgia. One paradigm shift into deep learning in 2012 and 10 years later we have something as general as GPT-4. Gatekeeping in academia is a real thing. Sad but true.

It is not ideas are easy, it’s just they are cheap. At a given time, multiple people may be having the same idea. So, only the speed of execution differs. I have seen horrible ideas get leverage because of execution.

You can see this prevalent in indie game industry. If you talk with designer’s everyone has their next big idea that can change gaming. But the industry is so unforgiving, any bad implementation or execution is severely punished. For example, see Dwarf Fortress. Stellar idea, awesome execution, but it took them really long time to hit mainstream. It is because the graphics wasn’t good enough for most mainstream people. Likewise, you see something like cookie-clicker going viral.

There is no not one single dimension to starting up. You need to be furious in all matters. Ideas are just multiplier’s. Good idea will give you enormous leverage but poor implementation is instant death. Even horrible idea when well executed can bring values.

It might be case they failed because they are MiT grads who most likely have a huge ego problem.

Out of all the clones, this one seems to be completely new. Eth else is frontend, but this, just shows the power. We all have our own Jarvis now.

It's difficult for starting people to ask out of school because their curiosity has been butchered by ego-maniac professors. For every 1 good professor, there are 1000 ego maniacs (professor + their TAs).

It hits the spot!

Learning quickly only works when every situation favors you. Because learning requires time while knowing doesn't. In this context, to be learned means to craft on your own. And only by knowing things do you get to master things.

If the above lines made sense to you, then only read the following paragraphs because I don't know if you can relate to this or not.

Learning quickly happens when you can know things quickly. That way you will have time to explore a lot and test a lot. Going through tutorials and documentation just allows you to know things. Therefore the quality of tutorials or mentors helps to grasp quickly because they just tell you what to know.

Real learning happens after you know things. Throughout my life, I got confused between knowing and learning. I can read or hear facts such that it allows me to recall the fact. But there were cases where I actually didn't understand the fact as I hadn't absorbed it completely. I just knew how things were but not the idea and reason why they were such.

When you want to truly learn something, to the finest detail, it really really requires a lot of effort and dedication. Picking up new skills is relatively easy. It just requires knowing enough to go through and solve your problems. But honing the skill to master level is really difficult.

Think about drawing sketches. You can just watch a tutorial on it and draw. Then close the tutorial and repeat it. You now know how to create at that pose. But when you want to change it, you will have to know about shading, pose, color, etc. Even if you know them, it's hard to use them because you may have never learned about them. Because to learn about colors, pose, shading requires experience and time. The more quickly you can iterate through it, the better you get about your preferences and style. Knowing them will only give you a shortcut. Learning them will give you a lot of creative freedom.